XPath Tester

XPath 1.0 section 1 says an expression evaluates to one of exactly four things: a node-set, a boolean, a number or a string. Your browser has a real XPath 1.0 engine, and it is what this page uses, so the answers are correct by construction. What a tester has to get right is ASKING: the browser API takes the result type you WANT as an argument and refuses the expression if it produces something else.

This page asks for whatever the expression produces and reports which of the four came back. So `count(//book)` gives a number, `string(//title)` gives a string, `count(//book) > 2` gives a boolean, and `//book` gives a node-set, and the type is named on screen beside the count.

Namespaces are the other half. The browser API takes a namespace resolver, and without one a prefixed name in an expression cannot be resolved at all. Worse, an unprefixed name in XPath 1.0 means "in no namespace", so `//title` selects nothing in an Atom feed, which is a document full of title elements. Every namespace the document declares is read out of it and bound here, and a default namespace is given a usable prefix so it can be written; you can add a binding of your own too. When an expression is valid and selects nothing, the page says whether a namespace is the reason.

Every node in a result carries an XPath that selects exactly that node, built by numbering each step among its same-named siblings, so you can copy one and use it directly. Attributes, text nodes, comments, CDATA sections and processing instructions are all shown with the kind of node they are.

A document that is not well formed gets a line, a column, the offending line with a caret under it, and a next step, from the same checker the XML formatter uses, which agrees with the browser on every document in that tool's corpus. An expression that does not compile is reported with the browser's own message plus a next step where the shape of the expression makes one obvious: unbalanced brackets, an unclosed string, or a prefix nothing binds.